RFLPtools (version 2.0)

RFLPdist2: Compute distances for RFLP data.

Description

If gel image quality is low, faint bands may be disregarded and may lead to wrong conclusions. This function computes the distance between the molecular weights of RFLP samples, including samples containing one or more additional bands. Thus, failures during band detection could be identified. Visualisation of band patterns using this method can be done by RFLPplot using the argument nrMissing.

Usage

RFLPdist2(x, distfun = dist, nrBands, nrMissing, LOD = 0,
          diag = FALSE, upper = FALSE)

Arguments

x

data.frame with RFLP data; see RFLPdata.

distfun

function computing the distance with default dist; cf. dist.

nrBands

samples with number of bands equal to nrBands are to be considered.

nrMissing

number of bands that might be missing.

LOD

threshold for low-bp bands.

diag

see dist

upper

see dist

Value

An object of class "dist" returned; cf. dist.

Details

For a given number of bands the given distance between the molecular weights is computed. It is assumed that a number of bands might be missing. Hence all samples with number of bands in nrBands, nrBands+1, ..., nrBands+nrMissing are compared.

If LOD > 0 is specified, it is assumed that missing bands can only occur for molecular weights smaller than LOD. As a consequence only samples which have nrBands bands with molecular weight larger or equal to LOD are selected.

For computing the distance between the molecular weight of a sample S1 with x bands and a Sample S2 with x+y bands the distances between the molecular weight of sample S1 and the molecular weight of all possible subsets of S2 with x bands are computed. The distance between S1 and S2 is then defined as the minimum of all these distances.

If LOD > 0 is specified, only all combinations of values below LOD are considered.

This option may be useful, if gel image quality is low, and the detection of bands is doubtful.

References

Flessa, F., Kehl, A., Kohl, M. Analysing diversity and community structures using PCR-RFLP: a new software application. Molecular Ecology Resources 2013 Jul; 13(4):726-33.

Ian A. Dickie, Peter G. Avis, David J. McLaughlin, Peter B. Reich. Good-Enough RFLP Matcher (GERM) program. Mycorrhiza 2003, 13:171-172.

See Also

RFLPdata, nrBands, RFLPdist, dist

Examples

# NOT RUN {
## Euclidean distance
data(RFLPdata)
nrBands(RFLPdata)
res0 <- RFLPdist(RFLPdata, nrBands = 4)
res1 <- RFLPdist2(RFLPdata, nrBands = 4, nrMissing = 1)
res2 <- RFLPdist2(RFLPdata, nrBands = 4, nrMissing = 2)
res3 <- RFLPdist2(RFLPdata, nrBands = 4, nrMissing = 3)

## assume missing bands only below LOD
res1.lod <- RFLPdist2(RFLPdata, nrBands = 4, nrMissing = 1, LOD = 60)

## hierarchical clustering
par(mfrow = c(2,2))
plot(hclust(res0), main = "0 bands missing")
plot(hclust(res1), main = "1 band missing")
plot(hclust(res2), main = "2 bands missing")
plot(hclust(res3), main = "3 bands missing")

## missing bands only below LOD
par(mfrow = c(1,2))
plot(hclust(res0), main = "0 bands missing")
plot(hclust(res1.lod), main = "1 band missing below LOD")

## Similarity matrix
library(MKomics)
myCol <- colorRampPalette(brewer.pal(8, "RdYlGn"))(128)
ord <- order.dendrogram(as.dendrogram(hclust(res1)))
temp <- as.matrix(res1)
simPlot(temp[ord,ord], col = rev(myCol), minVal = 0, 
        labels = colnames(temp), title = "(Dis-)Similarity Plot")

## missing bands only below LOD
ord <- order.dendrogram(as.dendrogram(hclust(res1.lod)))
temp <- as.matrix(res1.lod)
simPlot(temp[ord,ord], col = rev(myCol), minVal = 0, 
        labels = colnames(temp), title = "(Dis-)Similarity Plot\n1 band missing below LOD")


## or
library(lattice)
levelplot(temp[ord,ord], col.regions = rev(myCol),
          at = do.breaks(c(0, max(temp)), 128),
          xlab = "", ylab = "",
          ## Rotate label of x axis
          scales = list(x = list(rot = 90)),
          main = "(Dis-)Similarity Plot")


## Other distances
res11 <- RFLPdist2(RFLPdata, distfun = function(x) dist(x, method = "manhattan"),
                 nrBands = 4, nrMissing = 1)
res12 <- RFLPdist2(RFLPdata, distfun = corDist, nrBands = 4, nrMissing = 1)
res13 <- RFLPdist2(RFLPdata, distfun = corDist, nrBands = 4, nrMissing = 1, LOD = 60)
par(mfrow = c(2,2))
plot(hclust(res1), main = "Euclidean distance\n1 band missing")
plot(hclust(res11), main = "Manhattan distance\n1 band missing")
plot(hclust(res12), main = "Pearson correlation distance\n1 band missing")
plot(hclust(res13), main = "Pearson correlation distance\n1 band missing below LOD")
# }